UC Irvine

Macroscopic neural activity measured using fMRI, MEG and EEG shows consistent statistical dependencies between different areas of the brain which is labeled as functional connectivity. A major research direction in systems neuroscience has been the effort to trace functional connectivity back to structural connectivity, defined by the white-matter tracts of the brain. In this dissertation I examine the extent to which the structural connectivity constrains dynamic neural function. We first examined, in simulation and empirically, resting state MEG data from healthy subjects and tested whether effective connectivity estimates (models of functional connectivity) are sensitive to anatomical connectivity. Effective connectivity (EC) estimated using a graphical model assesses direct connectivity potentially providing a metric appropriate to compare against anatomical connections. In silico, we compared two different models for effective connectivity - complex valued Gaussian graphical models (cGGM) that emphasizes direct connectivity on the timescale of milliseconds and amplitude Gaussian Graphical models (aGGM) that emphasizes direct connectivity on the timescale of seconds. We found that the partial coherence, a summary statistic derived from cGGM, is more aligned to the anatomy than partial correlation (derived from aGGM). Using resting state MEG data, we found that partial coherence at higher frequencies (>30Hz) indeed provided the greatest alignment to the structural connectivity. However, at lower frequencies (1-30 Hz), we found that the partial correlation was more tightly coupled to the anatomy. We then examined the relationship between structure and function in a subject population with heterogeneous structural connections - subacute (< 1 month post stroke) patients of stroke who have lesions in the brain. Using fMRI time series parcelled into 114 brain areas, we estimated Gaussian Graphical Models (GGM) to derive the partial correlation and used virtual tractography for the structural connectivity. We found that partial correlations (relative to traditional correlations) strongly reflected the presence of structural connectivity damage. Further, applying voxel-based analyses driven by the streamlines (representing axonal fibers) allowed us to see modulation from damage across more functional connections suggesting parceling the brain inappropriately spatial smooths diverse neural activity. Finally, we examined the impact of structural connectivity damage on function in subacute and chronic (> 2 months post stroke) patients of stroke using EEG. We found increased structural damage was linked to increased overall connectivity (partial coherence) in both groups. Overall our results indicate that using simple models of effective connectivity (GGMs) allow better assessment of the relationship between macroscale structure and function but even so, structure mediates function only to a limited extent, with the relationship depending strongly on the timescale and spatial resolution of the effective connectivity.